brainstorming

SkillAI & models

Use when creating a feature/component or adding functionality. Fires BEFORE APEX Analyze to refine requirements via structured questioning.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the brainstorming skill

What this skill tells your AI

The instructions your AI receives, as published by fusengine/agents in plugins/ai-pilot/skills/brainstorming/SKILL.md and read by ahel’s review.

Skip it entirely for trivial fixes, typos, or simple renames -- those go straight to APEX.

Brainstorming Skill

Design-first approach: no code before design approval.


Agent Workflow (MANDATORY)

Before ANY brainstorming session, spawn agents in parallel, one Agent call each with a name:

  1. fuse-ai-pilot:explore-codebase - Understand project context, patterns, constraints
  2. fuse-ai-pilot:research-expert - Fetch best practices and documentation

After design approval, transition to APEX Analyze phase, passing along the explore-codebase and research-expert findings in the handoff (not just the design doc). APEX Analyze re-runs those same agents by default — carrying forward what was already gathered here avoids re-doing the same research twice.


Overview

ScenarioAction
New featureFull brainstorming (6 steps)
Major changeFull brainstorming (6 steps)
Component creationFull brainstorming (6 steps)
Trivial fix (1-3 lines)Skip brainstorming, go to APEX
Simple rename/typoSkip brainstorming, go to APEX

6-Step Process

Step 1: Explore Project Context

Gather context before asking questions:

  • git log --oneline -20 - Recent changes and direction
  • Existing code patterns and conventions
  • Related documentation and prior decisions
  • Tech stack and dependency constraints

Step 2: Ask Clarifying Questions (ONE AT A TIME)

Ask focused questions sequentially. Wait for each answer before the next.

Categories: purpose, constraints, success criteria, users, integrations.

See workflow.md for question categories

Step 3: Diverge, Then Converge to 2-3 Approaches

Generate ≥6-8 distinct approaches via a named technique (SCAMPER / reverse-brainstorming / analogies), judgment suspended — not 3 sizes of the same idea. Only then converge to 2-3 with trade-offs in table format:

ApproachProsConsRecommendation
Option A.........
Option B.........
Option C.........

Always include a clear recommendation with rationale.

Step 4: Present Design for Approval

Break design into digestible sections:

  1. Architecture overview
  2. Key components and responsibilities
  3. Data flow and state management
  4. Edge cases and error handling

Wait for explicit user approval before proceeding.

Step 5: Save Design Document

Save approved design to: docs/plans/YYYY-MM-DD-<topic>-design.md

Step 6: Transition to APEX

Hand off to APEX Analyze phase with the approved design as input.


Reference Guide

TopicReferenceWhen to Consult
Workflowworkflow.mdQuestion categories, proposal format, design template
Anti-Patternsanti-patterns.mdCatching rationalizations to skip brainstorming

Quick Reference

1. Explore   → git log, codebase, docs (agents in parallel)
2. Question  → ONE AT A TIME, wait for answers
3. Diverge   → 6-8 options (named technique), converge to 2-3 with trade-offs table
4. Design    → Present sections, get approval
5. Save      → docs/plans/YYYY-MM-DD-<topic>-design.md
6. Handoff   → APEX Analyze with approved design + prior research (no re-research)

Critical Rules

  1. NEVER write code before design approval - Design first, always
  2. Ask questions ONE AT A TIME - Never dump a list of 10 questions
  3. Always propose alternatives - Minimum 2 approaches with trade-offs
  4. Save the design doc - Creates audit trail and shared reference
  5. Get explicit approval - "Looks good" or similar before proceeding

Signals

GitHub stars
25
Forks
4
Last commit
Sep 2026
Hacker News mentions
14
Advanced
Catalog kind
skill
Gateway key
brainstorming-fusengine
Source
github.com/fusengine/agents